Abstract
J. Benesty et al. indicated that a method of a direct generalization of the affine projection algorithm (APA) to a multichannel case converges very slowly because the cross-correlation between input signals to all channels was not considered in it. MC-APA has been proposed to resolve this drawback. However, MC-APA carries out the update of coefficients of one channel by utilizing the properties of the input signal of the other channels, but not that of all channels. To overcome the defect of MC-APA, we proposed a new method in which all channel input signals are utilized for the coefficient correction of one channel. We show by many numerical examples that the proposed method have good convergence characteristics.
